Machine learning derived input-function in a dynamic 18F-FDG PET study of mice

Machine learning derived input-function in a dynamic 18F-FDG PET study of mice. Kuttner, S., Wickstrøm, K. K., Kalda, G., Dorraji, S E., Martin-Armas, M., Oteiza, A., Jenssen, R., Fenton, K., Sundset, R., & Axelsson, J. Biomedical Physics & Engineering Express, 6(1):015020, IOP Publishing, jan, 2020.

Paper doi abstract bibtex

Tracer kinetic modelling, based on dynamic 18F-fluorodeoxyglucose (FDG) positron emission tomography (PET) is used to quantify glucose metabolism in humans and animals. Knowledge of the arterial input-function (AIF) is required for such measurements. Our aim was to explore two non-invasive machine learning-based models, for AIF prediction in a small-animal dynamic FDG PET study. 7 tissue regions were delineated in images from 68 FDG PET/computed tomography mouse scans. Two machine learning-based models were trained for AIF prediction, based on Gaussian processes (GP) and a long short-term memory (LSTM) recurrent neural network, respectively. Because blood data were unavailable, a reference AIF was formed by fitting an established AIF model to vena cava and left ventricle image data. The predicted and reference AIFs were compared by the area under curve (AUC) and root mean square error (RMSE). Net-influx rate constants, Ki, were calculated with a two-tissue compartment model, using both predicted and reference AIFs for three tissue regions in each mouse scan, and compared by means of error, ratio, correlation coefficient, P value and Bland-Altman analysis. The impact of different tissue regions on AIF prediction was evaluated by training a GP and an LSTM model on subsets of tissue regions, and calculating the RMSE between the reference and the predicted AIF curve. Both models generated AIFs with AUCs similar to reference. The LSTM models resulted in lower AIF RMSE, compared to GP. Ki from both models agreed well with reference values, with no significant differences. Myocardium was highlighted as important for AIF prediction, but AIFs with similar RMSE were obtained also without myocardium in the input data. Machine learning can be used for accurate and non-invasive prediction of an image-derived reference AIF in FDG studies of mice. We recommend the LSTM approach, as this model predicts AIFs with lower errors, compared to GP.

@article{Kuttner_2020,
	doi = {10.1088/2057-1976/ab6496},
	url = {https://doi.org/10.1088/2057-1976/ab6496},
	year = 2020,
	month = {jan},
	publisher = {{IOP} Publishing},
	volume = {6},
	number = {1},
	pages = {015020},
	author = {Samuel Kuttner and Kristoffer Knutsen Wickstrøm and Gustav Kalda and S Esmaeil Dorraji and Montserrat Martin-Armas and Ana Oteiza and Robert Jenssen and Kristin Fenton and Rune Sundset and Jan Axelsson},
	title = {Machine learning derived input-function in a dynamic 18F-{FDG} {PET} study of mice},
	journal = {Biomedical Physics & Engineering Express},
	abstract = {Tracer kinetic modelling, based on dynamic 18F-fluorodeoxyglucose (FDG) positron emission tomography (PET) is used to quantify glucose metabolism in humans and animals. Knowledge of the arterial input-function (AIF) is required for such measurements. Our aim was to explore two non-invasive machine learning-based models, for AIF prediction in a small-animal dynamic FDG PET study. 7 tissue regions were delineated in images from 68 FDG PET/computed tomography mouse scans. Two machine learning-based models were trained for AIF prediction, based on Gaussian processes (GP) and a long short-term memory (LSTM) recurrent neural network, respectively. Because blood data were unavailable, a reference AIF was formed by fitting an established AIF model to vena cava and left ventricle image data. The predicted and reference AIFs were compared by the area under curve (AUC) and root mean square error (RMSE). Net-influx rate constants, Ki, were calculated with a two-tissue compartment model, using both predicted and reference AIFs for three tissue regions in each mouse scan, and compared by means of error, ratio, correlation coefficient, P value and Bland-Altman analysis. The impact of different tissue regions on AIF prediction was evaluated by training a GP and an LSTM model on subsets of tissue regions, and calculating the RMSE between the reference and the predicted AIF curve. Both models generated AIFs with AUCs similar to reference. The LSTM models resulted in lower AIF RMSE, compared to GP. Ki from both models agreed well with reference values, with no significant differences. Myocardium was highlighted as important for AIF prediction, but AIFs with similar RMSE were obtained also without myocardium in the input data. Machine learning can be used for accurate and non-invasive prediction of an image-derived reference AIF in FDG studies of mice. We recommend the LSTM approach, as this model predicts AIFs with lower errors, compared to GP.}
}

Downloads: 0

{"_id":"tJQ7iEQSxDYnwXoag","bibbaseid":"kuttner-wickstrm-kalda-dorraji-martinarmas-oteiza-jenssen-fenton-etal-machinelearningderivedinputfunctioninadynamic18ffdgpetstudyofmice-2020","authorIDs":[],"author_short":["Kuttner, S.","Wickstrøm, K. K.","Kalda, G.","Dorraji, S E.","Martin-Armas, M.","Oteiza, A.","Jenssen, R.","Fenton, K.","Sundset, R.","Axelsson, J."],"bibdata":{"bibtype":"article","type":"article","doi":"10.1088/2057-1976/ab6496","url":"https://doi.org/10.1088/2057-1976/ab6496","year":"2020","month":"jan","publisher":"IOP Publishing","volume":"6","number":"1","pages":"015020","author":[{"firstnames":["Samuel"],"propositions":[],"lastnames":["Kuttner"],"suffixes":[]},{"firstnames":["Kristoffer","Knutsen"],"propositions":[],"lastnames":["Wickstrøm"],"suffixes":[]},{"firstnames":["Gustav"],"propositions":[],"lastnames":["Kalda"],"suffixes":[]},{"firstnames":["S","Esmaeil"],"propositions":[],"lastnames":["Dorraji"],"suffixes":[]},{"firstnames":["Montserrat"],"propositions":[],"lastnames":["Martin-Armas"],"suffixes":[]},{"firstnames":["Ana"],"propositions":[],"lastnames":["Oteiza"],"suffixes":[]},{"firstnames":["Robert"],"propositions":[],"lastnames":["Jenssen"],"suffixes":[]},{"firstnames":["Kristin"],"propositions":[],"lastnames":["Fenton"],"suffixes":[]},{"firstnames":["Rune"],"propositions":[],"lastnames":["Sundset"],"suffixes":[]},{"firstnames":["Jan"],"propositions":[],"lastnames":["Axelsson"],"suffixes":[]}],"title":"Machine learning derived input-function in a dynamic 18F-FDG PET study of mice","journal":"Biomedical Physics & Engineering Express","abstract":"Tracer kinetic modelling, based on dynamic 18F-fluorodeoxyglucose (FDG) positron emission tomography (PET) is used to quantify glucose metabolism in humans and animals. Knowledge of the arterial input-function (AIF) is required for such measurements. Our aim was to explore two non-invasive machine learning-based models, for AIF prediction in a small-animal dynamic FDG PET study. 7 tissue regions were delineated in images from 68 FDG PET/computed tomography mouse scans. Two machine learning-based models were trained for AIF prediction, based on Gaussian processes (GP) and a long short-term memory (LSTM) recurrent neural network, respectively. Because blood data were unavailable, a reference AIF was formed by fitting an established AIF model to vena cava and left ventricle image data. The predicted and reference AIFs were compared by the area under curve (AUC) and root mean square error (RMSE). Net-influx rate constants, Ki, were calculated with a two-tissue compartment model, using both predicted and reference AIFs for three tissue regions in each mouse scan, and compared by means of error, ratio, correlation coefficient, P value and Bland-Altman analysis. The impact of different tissue regions on AIF prediction was evaluated by training a GP and an LSTM model on subsets of tissue regions, and calculating the RMSE between the reference and the predicted AIF curve. Both models generated AIFs with AUCs similar to reference. The LSTM models resulted in lower AIF RMSE, compared to GP. Ki from both models agreed well with reference values, with no significant differences. Myocardium was highlighted as important for AIF prediction, but AIFs with similar RMSE were obtained also without myocardium in the input data. Machine learning can be used for accurate and non-invasive prediction of an image-derived reference AIF in FDG studies of mice. We recommend the LSTM approach, as this model predicts AIFs with lower errors, compared to GP.","bibtex":"@article{Kuttner_2020,\n\tdoi = {10.1088/2057-1976/ab6496},\n\turl = {https://doi.org/10.1088/2057-1976/ab6496},\n\tyear = 2020,\n\tmonth = {jan},\n\tpublisher = {{IOP} Publishing},\n\tvolume = {6},\n\tnumber = {1},\n\tpages = {015020},\n\tauthor = {Samuel Kuttner and Kristoffer Knutsen Wickstrøm and Gustav Kalda and S Esmaeil Dorraji and Montserrat Martin-Armas and Ana Oteiza and Robert Jenssen and Kristin Fenton and Rune Sundset and Jan Axelsson},\n\ttitle = {Machine learning derived input-function in a dynamic 18F-{FDG} {PET} study of mice},\n\tjournal = {Biomedical Physics & Engineering Express},\r\n\tabstract = {Tracer kinetic modelling, based on dynamic 18F-fluorodeoxyglucose (FDG) positron emission tomography (PET) is used to quantify glucose metabolism in humans and animals. Knowledge of the arterial input-function (AIF) is required for such measurements. Our aim was to explore two non-invasive machine learning-based models, for AIF prediction in a small-animal dynamic FDG PET study. 7 tissue regions were delineated in images from 68 FDG PET/computed tomography mouse scans. Two machine learning-based models were trained for AIF prediction, based on Gaussian processes (GP) and a long short-term memory (LSTM) recurrent neural network, respectively. Because blood data were unavailable, a reference AIF was formed by fitting an established AIF model to vena cava and left ventricle image data. The predicted and reference AIFs were compared by the area under curve (AUC) and root mean square error (RMSE). Net-influx rate constants, Ki, were calculated with a two-tissue compartment model, using both predicted and reference AIFs for three tissue regions in each mouse scan, and compared by means of error, ratio, correlation coefficient, P value and Bland-Altman analysis. The impact of different tissue regions on AIF prediction was evaluated by training a GP and an LSTM model on subsets of tissue regions, and calculating the RMSE between the reference and the predicted AIF curve. Both models generated AIFs with AUCs similar to reference. The LSTM models resulted in lower AIF RMSE, compared to GP. Ki from both models agreed well with reference values, with no significant differences. Myocardium was highlighted as important for AIF prediction, but AIFs with similar RMSE were obtained also without myocardium in the input data. Machine learning can be used for accurate and non-invasive prediction of an image-derived reference AIF in FDG studies of mice. We recommend the LSTM approach, as this model predicts AIFs with lower errors, compared to GP.}\r\n}\n\n","author_short":["Kuttner, S.","Wickstrøm, K. K.","Kalda, G.","Dorraji, S E.","Martin-Armas, M.","Oteiza, A.","Jenssen, R.","Fenton, K.","Sundset, R.","Axelsson, J."],"key":"Kuttner_2020","id":"Kuttner_2020","bibbaseid":"kuttner-wickstrm-kalda-dorraji-martinarmas-oteiza-jenssen-fenton-etal-machinelearningderivedinputfunctioninadynamic18ffdgpetstudyofmice-2020","role":"author","urls":{"Paper":"https://doi.org/10.1088/2057-1976/ab6496"},"metadata":{"authorlinks":{}},"downloads":0},"bibtype":"article","biburl":"https://drive.google.com/uc?id=1npQd_kFQJ1WhoHFdMxm9f24VTXXUJ6kw","creationDate":"2020-06-04T10:58:19.190Z","downloads":0,"keywords":[],"search_terms":["machine","learning","derived","input","function","dynamic","18f","fdg","pet","study","mice","kuttner","wickstrøm","kalda","dorraji","martin-armas","oteiza","jenssen","fenton","sundset","axelsson"],"title":"Machine learning derived input-function in a dynamic 18F-FDG PET study of mice","year":2020,"dataSources":["oPYGjz5LJw7BLCPiK","c6fLyraeGp89Bmhyo","qb42Xqdsuo5n9MtiJ","3HekLNYCKYdmLkjZH","XhEeRFbisz5SLRLws","EoffqxLuQbhKRpNmQ","P5LkHJDQ7GXibcCon"]}